A surveillance activity recognition model based on hidden Markov model
Liang Hao-zhe, Kuihua Huang, Guohui Li · 2012
In this paper a novel activity recognition model based on Hidden Markov model was proposed. For the HMM parameters learning problem, a two-phase model including a bottom-up process and a top-down process was introduced. Bottom-up used Dirichlet Mixture Model to learn the HMM structure automatically and top-down defined a generative clustering process, which was called HMM-mixture. Both processes were unsupervised. The performance of the proposed model was tested by real surveillance video and an application for classification of activity was also showed. Clusters of HMM-trajectory were successfully recognized by the proposed model and properly classification results were achieved.